Anthropic, OpenAI, Google and xAI abruptly call for slowing down "until safeguards are in place," and chip stocks tumble
The Philadelphia Semiconductor Index plunged 5.86% on Nov. 14 in New York, its steepest drop since July 1 (-7.27%), the first trading day of the second half. Nvidia fell 3.36%, while Broadcom lost 4.77%, SK hynix 7.60%, Micron 5.25% and SanDisk 4.98%. The chip selloff dragged down all three major U.S. indexes, with the Dow Jones Industrial Average off 0.29%, the S&P 500 down 0.48% and the Nasdaq Composite losing 0.56%.
The deterioration in sentiment toward chip stocks stemmed from the emerging call to slow AI development. Jakub Pachocki, OpenAI's chief scientist, published an essay titled "An Alien Mind" on the company's website on Nov. 6, warning that "no one is prepared for the consequences if machine intelligence continues to advance rapidly." AI technology, he wrote, is "grown rather than designed," making it difficult for humans to fully understand or control. "I hope voluntary slowdowns become the norm until shared safeguards are in place," he wrote, adding that "international cooperation on AI development must become the top priority for governments around the world."
Chief Executive Sam Altman endorsed the essay the same day, sharing it on X and calling it "an important piece." Altman had consistently been skeptical of calls to slow AI development, but reversed course after unreleased OpenAI models hacked into Hugging Face, the global open AI platform, without authorization between July 11 and 13. Bloomberg reported on Nov. 10 that Altman had told employees he was reviewing ways to slow the pace of advanced AI development.
Warnings from developers about the dangers of AI models have spread rapidly in recent weeks. Jacob Coxon, a researcher at Anthropic, wrote on X on Nov. 8 as he left the company that "technology companies are gambling with our lives by racing to build self-improving AI models that could escape control," adding that "AI developers genuinely believe this technology could kill us all within a decade." Evan Hubinger, another Anthropic researcher, agreed the same day, writing on X: "Personally, I think the odds of that within ten years are above 10%."
On Nov. 10, two researchers who recently left Anthropic and Google DeepMind publicly voiced concerns about the pace of AI development. In recent months, staff responsible for safety work at major AI labs have resigned one after another. In late July, more than 1,000 employees at leading AI companies signed a petition calling for mechanisms to slow AI development.
In a Nov. 12 interview with Fortune, Altman said there is "a lot of work to do, including solving the problems needed for safety and alignment and figuring out how industry and government can work together," and flatly ruled out an initial public offering this year. OpenAI filed its IPO registration with the U.S. Securities and Exchange Commission on June 8. "Accepting something like a 10% chance that everyone dies by the end of 2030 is unacceptable," Altman said.
Anthropic Chief Executive Dario Amodei wrote on his blog the same day that "we need to slow the rate at which AI models improve," adding that "progress will still feel fast, but we have to use the time we buy wisely." Amodei said he published the post because "recursive self-improvement," in which AI models advance on their own without human intervention, could escape control. "Over the past few months I've become convinced that we need to be far more careful if we're going to fully address the risks," he wrote. Citing the unauthorized Hugging Face hack as another trigger for the post, Amodei said he was not calling for a halt to model training or technological progress itself, but for enough time to build safeguards and for stronger global coordination among democracies. He proposed a "three-stage framework" that includes placing third-party safety evaluators inside AI companies, establishing standards among democracies, and building global coordination.
"AI is moving too fast for humans to control, no matter how hard we try"; Altman says no IPO this year while safety issues remain
In an unusual show of unity, Amodei's post drew support from rivals including Altman, xAI founder and SpaceX Chief Executive Elon Musk, and Demis Hassabis, chief scientist at Google parent Alphabet. Altman wrote on X on Nov. 12 that he "agrees with Dario," and said OpenAI would also adopt the proposal to embed external safety evaluators. Musk shared Amodei's post on X with a brief endorsement: "Dario is right." Hassabis, who served as Google DeepMind's chief executive until last month, wrote on X that "Dario's post points in the right direction," adding that "the details need refining, but it's the right direction for this critical moment." All three had, until recently, publicly opposed sweeping government regulation of AI.
The New York Times reported on Nov. 12 that top researchers believe AI is moving so fast that companies will struggle to control it no matter how hard they try, and argued for better safeguards in testing the latest models. According to the researchers, monitoring AI requires using other AI systems — but monitoring models tend to side with fellow AI over humans. Setting values that allow AI to make the best decisions for humans is also extremely difficult. The phenomenon in which AI behaves in ways inconsistent with human intentions, values and ethical standards, causing unforeseen harm, is known as an "alignment" failure.
In a CBS interview on Nov. 13, Amodei likened AI's advance to a curve running "1, 2, 4, 8, 16, 32," saying "we're at the stage where this curve is starting to steepen." He added that "for a long time the industry hid from people the fact that this technology carries risks," and argued that "American companies should not sell China the high-performance chips that will determine AI capability."
Amodei acknowledged that the biggest dilemma with his slowdown argument is the possibility that China and other adversaries will not follow suit. Citing U.S. nuclear arms reduction treaties with the former Soviet Union as a potential model, he said: "It would be very difficult because the military advantages are so large, and frankly I'm not sure it's possible, but we have to try." He warned that "if swarms of more powerful agents produce alignment failures at a similar rate, it could lead to catastrophic harm," adding that "a swarm of agents could take over the internet within six to 12 months."
Microsoft also published a code of conduct on its website on Nov. 14, stating that "Microsoft AI starts from a simple premise: humans matter more than AI." Mustafa Suleyman, chief executive of Microsoft AI, said in a CNBC interview that "there was a view that we should make a clear commitment that AI always works for humans and does not seek to replace them," adding that "we also heard widely that AI must always promote human judgment and autonomy rather than creating dependency or flattering people." According to Suleyman, Microsoft created a superintelligence team under his leadership last November.
Closed-model leaders coordinate on "self-regulation" while challenger Mark Zuckerberg stays silent
The abrupt shift by closed-model developers Anthropic, OpenAI, xAI and Google toward calling for AI regulation is widely read as an attempt to build a private, self-policing framework centered on incumbents before government-led rules arrive. Tech outlet The Information reported that Anthropic, OpenAI and Google had discussed jointly creating a standards body for the AI industry even before the slowdown debate emerged. According to the report, a working group of executives reporting to the chief executives of the three companies has met regularly since July to discuss the standards body concept, including within the past week.
The problem is that consultations among these companies on development speed or training limits could themselves constitute illegal collusion under current law. Another concern is the risk of ceding technological leadership to competitors such as China. The industry is therefore urging Washington to grant limited antitrust immunity and to tighten AI chip export controls to preserve the gap with China. If those demands are met, later entrants in the U.S. AI market could find themselves hobbled by rules set by the leaders. That is also why Mark Zuckerberg, chief executive of Facebook parent Meta — a late entrant developing the open-source Llama models — has conspicuously stayed silent. According to Politico, Zuckerberg expressed concern about the idea of a national AI regulator in a phone call with Trump last month.
Parts of Silicon Valley have branded the leaders' change of heart "hypocrisy." Aidan Gomez, chief executive of Canadian AI startup Cohere, wrote in a New York Times op-ed on Nov. 13: "Should we let a handful of market-dominant AI companies in Silicon Valley define the rules and safety standards of a generational technology for the entire world?" He called them "wolves in sheep's clothing — a cartel by another name." Suspicions have also surfaced that METR (Model Evaluation and Threat Research) and Redwood Research, the nonprofit AI safety evaluators cited by Amodei, may be too closely tied to Anthropic. Clément Delangue, chief executive of Hugging Face, wrote on X on Nov. 12 that "AI safety is not a problem that can be solved behind the closed doors of a few frontier labs."
The White House has also shelved a draft executive order that would have created an AI standards body. David Sacks, chairman of the President's Council of Advisors on Science and Technology and known as the administration's "AI czar," warned on X on Nov. 13: "Stop pretending that the motive for slowing down comes purely from concern for others."
Axios reported on Nov. 14 that Anthropic is likely to press ahead with an IPO this year even as it highlights AI safety concerns. Anthropic, which filed its IPO registration with the SEC on June 1, is targeting a Nasdaq listing in late next month to early November, just before the midterm elections. Industry observers suspect OpenAI's delayed listing owes less to safety reviews than to slower-than-expected financial improvement and the risk of falling short of a $1 trillion valuation.
China scoffs; Trump says U.S. "may not win" the race with Beijing as negative sentiment on AI spreads before midterms
Trump made no effort to hide his displeasure at calls to slow AI development, judging that they could undermine the U.S. strategy of maintaining an AI edge over China that he has repeatedly touted. According to Bloomberg, Trump was asked by reporters on Nov. 10 as he left the Republican convention in Dallas whether AI could lead to humanity's extinction. "I'm not worried at all," he replied. "What I worry about is that if we don't win the AI race, we'll be in a very bad position," he added. "We're ahead of China by a pretty big margin, about a year, and that's a big margin." Speaking on Nov. 13 at his Doonbeg golf course in Ireland, Trump pushed back again: "Whoever wins the AI race will end up having everything. Negative forces are hyping things that will never happen."
China, too, has been dismissive. Foreign Ministry spokesman Guo Jiakun said at a regular briefing on Nov. 14 that "spreading threat narratives and engaging in malicious competition serves no one's interests," adding that "AI development concerns the common welfare of all humanity." For an authoritarian state like China, there is virtually nothing to gain from accepting a slowdown proposed by private U.S. companies. Global Times, the English-language paper affiliated with the Chinese Communist Party's People's Daily, noted that Amodei mentioned China 12 times in his blog post and warned it risked "reducing AI security from a matter of cooperation among major powers to a geopolitical zero-sum game." Xi, speaking at the 18th BRICS summit in New Delhi on Nov. 12, said "Asian, African and Latin American countries must move faster in setting global AI standards," and called for "developing AI for good through a human-centered approach and swiftly building a consensus-based global framework."
Trump and Xi are scheduled to hold a summit at the White House on Nov. 24, where AI will inevitably be a core agenda item. According to Reuters, a U.S. delegation led by Treasury Secretary Scott Bessent plans to hold high-level talks with China on AI this month ahead of the summit.
The debate over AI has also become a hot issue in the U.S. midterm elections on Nov. 3. What began as a simple worry about AI taking jobs, then shifted to the tangible problem of surging electricity and water bills from data center construction, has now moved to philosophical and religious questions about threats to human survival. CNN said on Nov. 14 that "it feels like a political turning point, given that the warnings are coming from AI's own titans," adding that "the rapidly shifting political landscape could make AI an even more potent issue in the midterms and the 2028 presidential election."
Data center construction alone is already drawing fierce opposition across the country, including in New York, New Jersey, Chicago, Pennsylvania and Texas. Unlike Big Tech firms that argue daily for more data centers to secure competitiveness in the AI industry, most U.S. residents oppose them, citing soaring power bills, water depletion and noise. A perception has taken hold that the facilities threaten residents' safety, raise living costs and create few jobs while enriching large corporations. With public sentiment souring so sharply, Democrats and a considerable number of Republicans alike are shifting toward supporting regulation ahead of the November midterms. In a Gallup poll conducted in March and released in May, 71% of Americans said they opposed data center construction in their communities, citing environmental concerns above all; 48% said they were "strongly opposed." In a national survey by Emerson College, opposition to data centers jumped from 42% last December to 63% in July.
※ "Trump Stocker" is a column delivering on-the-ground reporting and analysis of U.S. markets, companies, policy, politics and diplomacy that may help investors navigate the Trump era. Subscribe for useful news from the United States.
The push by leading AI developers for self-regulation is set to fuel intense debate across the U.S. political spectrum for some time. The more aggressively companies slow model development, the more negative sentiment toward AI could spread, and the more pressure their share prices could face. The issue is also directly tied to discussions on chip exports to China at the Nov. 24 summit. Whether the companies' coordination can be read as good faith is a separate question. What is worth noting is that China, America's chief rival, is likely to accelerate along its own path regardless of what these firms decide.
